Research On Image Deblurring Based On Nonlocal Regularization With A Non-Convex Sparsity Constraint

In modern society,with the rapid development of information technology,image has drawn more and more attention as an important information expression carrier.In many practical applications,the requirements for image clarity are often high.However,during the process of image acquisition,storage and transmission,the quality of images will be degraded to a certain degree due to various man-made and environmental factors,that is,image degradation.Degraded images bring great difficulties to researchers in the follow-up analysis and processing,therefore,the role of image restoration becomes especially important.The purpose of image restoration is to reconstruct the original clear image from its degraded version,which is an ill-posed inverse problem.Solving the ill-posed problem directly often cannot result in the desired result,and the regularization method is an effective way to solve this problem.Based on the nonlocal regularization technique,this paper focuses on the image deblurring problem in image restoration.By observing the probability distribution of the weighted gradient vector of the clear image on each dimension,we find that the generalized Gaussian distribution can be used to model it more accurately.Therefore,in this paper,the generalized Gaussian distribution is applied to the regularization item of the optimization problem,and the image deblurring method based on nonlocal regularization with a non-convex sparsity constraint is proposed.Compared with other state-of-the-art deblurring methods in three typical symmetric blur(including Gaussian blur,uniform blur and out-of-focus blur)cases,we find that the proposed method has obvious advantages in the deblurring performance in both subjective and objective evaluation,which further verifies the effectiveness of the proposed method.